CIF 3

External user-defined Java functions

External user-defined Java functions allow the use of static Java methods from CIF specifications. This page explains how to use such Java methods, how the simulator loads and invokes them, what is allowed, etc. It explicitly does not explain how to write or compile Java code.

External implementation reference

Consider the following examples of external user-defined Java function:

func real f1(real x): "java:java.lang.Math.expm1";

func int f2(bool x): "java:pkg.Cls.method|../some.jar;bin";

Two external user-defined functions are given as examples. Both have a string literal as external implementation reference. For Java functions, the string literal must start with java:. Then the absolute name of the Java method that is to be invoked is given. That is, the package name, class name, and method name are given, separated by dots (.).

Optionally, a Java class path may be given at the end, preceded by a bar (|) character. The class path entries must be separated by semicolons (;) regardless of the operating system. The class path entries are absolute or relative local file system paths to Jar files or directories with Java class files. Relative paths are resolved against the directory that contains the CIF specification in which the external user-defined function is declared. Both Linux/Mac path separators (/) and Windows path separators (\) may be used in class path entries. However, \ must be escaped as \\ in CIF string literals, and thus it is recommended to always use / as path separator. A discussion of the Java class path concept is beyond the scope of this documentation.

Supported types and values

Not all CIF data types are supported for the parameters and return values of external user-defined Java functions. The following table lists the allowed CIF types, and their corresponding Java types.

CIF type Java type
bool java.lang.Boolean / boolean
int java.lang.Integer / int
int[a..b] java.lang.Integer / int
real java.lang.Double / double
string java.lang.String
list t java.util.List<t>
set t java.util.List<t>
dict(k: v) java.util.Map<k, v>
tuple(t1, t2, ..., tn) java.util.List<java.lang.Object>

Unless otherwise specified, the primitive Java types (boolean, int, and double) can be used interchangeably with their class variants (java.lang.Boolean, java.lang.Integer, and java.lang.Double). For lists and sets, the element type recursively affects the type parameters of the Java types. Similarly, for dictionaries, the key and value types affect the type parameters of the Java Map type. Since Java doesn’t have tuples, and there is no way in Java to define a type with a variable number of type parameters, CIF tuples are passed to Java as a list of objects. It may be assumed that such lists have the same number of elements as there are fields in the original tuple, and that the elements have types that correspond to the types of the field of the tuple. The following table shows a few examples:

CIF type Java type
bool boolean
int int
list int java.util.List<java.lang.Integer>
set tuple(int a, b) java.util.Set<java.util.List<java.lang.Object>>

For parameters, no null values are ever passed, not even as elements of lists, etc. Furthermore, for real typed values, NaN, positive infinity, and negative infinity are never passed along.

For return values, null values, as well as NaN, positive infinity, and negative infinity are considered runtime errors. Furthermore, integer values outside their CIF integer type range bounds are considered runtime errors as well.

Method resolution

The class as specified in the external implementation reference is loaded using a Java class loader. By default, the system class loader is used. If additional class path entries are specified in the external implementation reference, an extended class loader is used that searches in the additional class path entries before deferring to the system class loader. The class is loaded using the binary name of the class.

The method that is resolved must be a static method. The access modifier of the static method is irrelevant. That is, it doesn’t matter whether the method is private, protected, public, etc.

The method is resolved in the Java Virtual Machine (JVM) that runs the simulator. The method is resolved by name, and parameter types. The parameter types only include the top level type. That is, for CIF type list int, the Java type java.util.List is used as the type of the parameter. If more than one method with the same parameter types is declared in the class, and one of these methods has a return type that is more specific than any of the others, that method is used. Otherwise one of the methods is chosen arbitrarily.

For parameters that have primitive Java types as their type, the primitive types are used to resolve the method instead of their classes. In other words, when resolving the method, or the proper overload of the method, int and java.lang.Integer are considered incompatible.

The CIF 3 simulator employs lazy loading. That is, if the external Java function is never called, it is also never loaded.

Method invocation

The method is invoked in the Java Virtual Machine (JVM) that runs the simulator, using reflection. By default, it is invoked asynchronously.

Since the parameter types are not checked recursively when the method is resolved, methods may fail to execute properly if the type signatures don’t match exactly. Similarly, for the return type only the top level type is checked. If a return value does not match the CIF type (wrong element type, wrong tuple length, etc), simulation will fail with a runtime error.

Any exceptions that occur during execution of the Java method are caught by the simulator. When the simulator catches such an exception, it will print the Java stack trace of the exception to the console, and simulation will be terminated.

Side effects

As all functions in CIF, external user-defined Java functions should not have any side effects. That is, if called with the same arguments, it should return the same value each time. More concretely, maintaining state information should be avoided, as should file I/O, and reading from stdin.

Writing to stdout and stderr should not affect the result of the method. It should however be avoided as well, as such writes are associated with the streams of the Java Virtual Machine (JVM), and not the simulator. The output will not end up on the Eclipse console of the simulator.

By default, the Java methods are invoked asynchronously, on a freshly created thread. That is, there is no guarantee that an invocation uses the same thread as the previous invocation. If static fields (class variables) are used to maintain state, this may cause problems if thread safety is not ensured by the Java method. If synchronous execution is used, all external function calls are invoked from the simulation thread, and such thread safety issues should not be a problem.

You should assume that each external function is loaded using a different class loader. As such, multiple Java methods from the same class, from the same external library, may be loaded using different class loaders as well. As such, shared static fields (class variables) of those classes can in general not be shared by different external functions.